Menu
Create excellent command-line tools. Use when asked to build CLIs, add commands/options, adjust input/output behavior, or update help text.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| name | cli-authoring |
| description | Create excellent command-line tools. Use when asked to build CLIs, add commands/options, adjust input/output behavior, or update help text. |
This document covers both high-level design philosophy, and concrete guidelines. Itâs heavier on the guidelines because our philosophy as practitioners is not to philosophize too much. We believe in learning by example, so weâve provided plenty of those.
This guide doesnât cover full-screen terminal programs like emacs and vim. Full-screen programs are niche projectsâvery few of us will ever be in the position to design one.
This guide is also agnostic about programming languages and tooling in general.
Who is this guide for?
These are what we consider to be the fundamental principles of good CLI design.
Traditionally, UNIX commands were written under the assumption they were going to be used primarily by other programs. They had more in common with functions in a programming language than with graphical applications.
Today, even though many CLI programs are used primarily (or even exclusively) by humans, a lot of their interaction design still carries the baggage of the past. Itâs time to shed some of this baggage: if a command is going to be used primarily by humans, it should be designed for humans first.
A core tenet of the original UNIX philosophy is the idea that small, simple programs with clean interfaces can be combined to build larger systems. Rather than stuff more and more features into those programs, you make programs that are modular enough to be recombined as needed.
In the old days, pipes and shell scripts played a crucial role in the process of composing programs together. Their role might have diminished with the rise of general-purpose interpreted languages, but they certainly havenât gone away. Whatâs more, large-scale automationâin the form of CI/CD, orchestration and configuration managementâhas flourished. Making programs composable is just as important as ever.
Fortunately, the long-established conventions of the UNIX environment, designed for this exact purpose, still help us today. Standard in/out/err, signals, exit codes and other mechanisms ensure that different programs click together nicely. Plain, line-based text is easy to pipe between commands. JSON, a much more recent invention, affords us more structure when we need it, and lets us more easily integrate command-line tools with the web.
Whatever software youâre building, you can be absolutely certain that people will use it in ways you didnât anticipate. Your software will become a part in a larger systemâyour only choice is over whether it will be a well-behaved part.
Most importantly, designing for composability does not need to be at odds with designing for humans first. Much of the advice in this document is about how to achieve both.
The terminalâs conventions are hardwired into our fingers. We had to pay an upfront cost by learning about command line syntax, flags, environment variables and so on, but it pays off in long-term efficiency⊠as long as programs are consistent.
Where possible, a CLI should follow patterns that already exist. Thatâs what makes CLIs intuitive and guessable; thatâs what makes users efficient.
That being said, sometimes consistency conflicts with ease of use. For example, many long-established UNIX commands don't output much information by default, which can cause confusion or worry for people less familiar with the command line.
When following convention would compromise a programâs usability, it might be time to break with itâbut such a decision should be made with care.
The terminal is a world of pure information. You could make an argument that information is the interfaceâand that, just like with any interface, thereâs often too much or too little of it.
A command is saying too little when it hangs for several minutes and the user starts to wonder if itâs broken. A command is saying too much when it dumps pages and pages of debugging output, drowning whatâs truly important in an ocean of loose detritus. The end result is the same: a lack of clarity, leaving the user confused and irritated.
It can be very difficult to get this balance right, but itâs absolutely crucial if software is to empower and serve its users.
When it comes to making functionality discoverable, GUIs have the upper hand. Everything you can do is laid out in front of you on the screen, so you can find what you need without having to learn anything, and perhaps even discover things you didnât know were possible.
It is assumed that command-line interfaces are the opposite of thisâthat you have to remember how to do everything. The original Macintosh Human Interface Guidelines, published in 1987, recommend âSee-and-point (instead of remember-and-type),â as if you could only choose one or the other.
These things neednât be mutually exclusive. The efficiency of using the command-line comes from remembering commands, but thereâs no reason the commands canât help you learn and remember.
Discoverable CLIs have comprehensive help texts, provide lots of examples, suggest what command to run next, suggest what to do when there is an error. There are lots of ideas that can be stolen from GUIs to make CLIs easier to learn and use, even for power users.
Citation: The Design of Everyday Things (Don Norman), Macintosh Human Interface Guidelines
GUI design, particularly in its early days, made heavy use of metaphor: desktops, files, folders, recycle bins. It made a lot of sense, because computers were still trying to bootstrap themselves into legitimacy. The ease of implementation of metaphors was one of the huge advantages GUIs wielded over CLIs. Ironically, though, the CLI has embodied an accidental metaphor all along: itâs a conversation.
Beyond the most utterly simple commands, running a program usually involves more than one invocation. Usually, this is because itâs hard to get it right the first time: the user types a command, gets an error, changes the command, gets a different error, and so on, until it works. This mode of learning through repeated failure is like a conversation the user is having with the program.
Trial-and-error isnât the only type of conversational interaction, though. There are others:
git adds, followed by a git commit).cd and ls to get a sense of a directory structure, or git log and git show to explore the history of a file.Acknowledging the conversational nature of command-line interaction means you can bring relevant techniques to bear on its design. You can suggest possible corrections when user input is invalid, you can make the intermediate state clear when the user is going through a multi-step process, you can confirm for them that everything looks good before they do something scary.
The user is conversing with your software, whether you intended it or not. At worst, itâs a hostile conversation which makes them feel stupid and resentful. At best, itâs a pleasant exchange that speeds them on their way with newfound knowledge and a feeling of achievement.
Further reading: The Anti-Mac User Interface (Don Gentner and Jakob Nielsen)
Robustness is both an objective and a subjective property. Software should be robust, of course: unexpected input should be handled gracefully, operations should be idempotent where possible, and so on. But it should also feel robust.
You want your software to feel like it isnât going to fall apart. You want it to feel immediate and responsive, as if it were a big mechanical machine, not a flimsy plastic âsoft switch.â
Subjective robustness requires attention to detail and thinking hard about what can go wrong. Itâs lots of little things: keeping the user informed about whatâs happening, explaining what common errors mean, not printing scary-looking stack traces.
As a general rule, robustness can also come from keeping it simple. Lots of special cases and complex code tend to make a program fragile.
Command-line tools are a programmerâs creative toolkit, so they should be enjoyable to use. This doesnât mean turning them into a video game, or using lots of emoji (though thereâs nothing inherently wrong with emoji đ). It means giving the user the feeling that you are on their side, that you want them to succeed, that you have thought carefully about their problems and how to solve them.
Thereâs no list of actions you can take that will ensure they feel this way, although we hope that following our advice will take you some of the way there. Delighting the user means exceeding their expectations at every turn, and that starts with empathy.
The world of the terminal is a mess. Inconsistencies are everywhere, slowing us down and making us second-guess ourselves.
Yet itâs undeniable that this chaos has been a source of power. The terminal, like the UNIX-descended computing environment in general, places very few constraints on what you can build. In that space, all manner of invention has bloomed.
Itâs ironic that this document implores you to follow existing patterns, right alongside advice that contradicts decades of command-line tradition. Weâre just as guilty of breaking the rules as anyone.
The time might come when you, too, have to break the rules. Do so with intention and clarity of purpose.
âAbandon a standard when it is demonstrably harmful to productivity or user satisfaction.â â Jef Raskin, The Humane Interface
This is a collection of specific things you can do to make your command-line program better.
The first section contains the essential things you need to follow. Get these wrong, and your program will be either hard to use or a bad CLI citizen.
The rest are nice-to-haves. If you have the time and energy to add these things, your program will be a lot better than the average program.
The idea is that, if you donât want to think too hard about the design of your program, you donât have to: just follow these rules and your program will probably be good. On the other hand, if youâve thought about it and determined that a rule is wrong for your program, thatâs fine. (Thereâs no central authority that will reject your program for not following arbitrary rules.)
Alsoâthese rules arenât written in stone. If you disagree with a general rule for good reason, we hope youâll propose a change.
There are a few basic rules you need to follow. Get these wrong, and your program will be either very hard to use, or flat-out broken.
Use a command-line argument parsing library where you can. Either your languageâs built-in one, or a good third-party one. They will normally handle arguments, flag parsing, help text, and even spelling suggestions in a sensible way.
Here are some that we like:
Return zero exit code on success, non-zero on failure. Exit codes are how scripts determine whether a program succeeded or failed, so you should report this correctly. Map the non-zero exit codes to the most important failure modes.
Send output to stdout.
The primary output for your command should go to stdout.
Anything that is machine readable should also go to stdoutâthis is where piping sends things by default.
Send messaging to stderr.
Log messages, errors, and so on should all be sent to stderr.
This means that when commands are piped together, these messages are displayed to the user and not fed into the next command.
Display extensive help text when asked.
Display help when passed -h or --help flags.
This also applies to subcommands which might have their own help text.
Display concise help text by default.
When myapp or myapp subcommand requires arguments to function,
and is run with no arguments,
display concise help text.
You can ignore this guideline
if your program is interactive by default (e.g. npm init).
The concise help text should only include:
--help flag for more information.jq does this well.
When you type jq, it displays an introductory description and an example, then prompts you to pass jq --help for the full listing of flags:
$ jq
jq - commandline JSON processor [version 1.6]
Usage: jq [options] <jq filter> [file...]
jq [options] --args <jq filter> [strings...]
jq [options] --jsonargs <jq filter> [JSON_TEXTS...]
jq is a tool for processing JSON inputs, applying the given filter to
its JSON text inputs and producing the filter's results as JSON on
standard output.
The simplest filter is ., which copies jq's input to its output
unmodified (except for formatting, but note that IEEE754 is used
for number representation internally, with all that that implies).
For more advanced filters see the jq(1) manpage ("man jq")
and/or https://stedolan.github.io/jq
Example:
$ echo '{"foo": 0}' | jq .
{
"foo": 0
}
For a listing of options, use jq --help.
Show full help when -h and --help is passed.
All of these should show help:
$ myapp
$ myapp --help
$ myapp -h
Ignore any other flags and arguments that are passedâyou should be able to add -h to the end of anything and it should show help.
Donât overload -h.
If your program is git-like, the following should also offer help:
$ myapp help
$ myapp help subcommand
$ myapp subcommand --help
$ myapp subcommand -h
Provide a support path for feedback and issues. A website or GitHub link in the top-level help text is common.
In help text, link to the web version of the documentation. If you have a specific page or anchor for a subcommand, link directly to that. This is particularly useful if there is more detailed documentation on the web, or further reading that might explain the behavior of something.
Lead with examples. Users tend to use examples over other forms of documentation, so show them first in the help page, particularly the common complex uses. If it helps explain what itâs doing and it isnât too long, show the actual output too.
You can tell a story with a series of examples, building your way toward complex uses.
If youâve got loads of examples, put them somewhere else, in a cheat sheet command or a web page. Itâs useful to have exhaustive, advanced examples, but you donât want to make your help text really long.
For more complex use cases, e.g. when integrating with another tool, it might be appropriate to write a fully-fledged tutorial.
Display the most common flags and commands at the start of the help text. Itâs fine to have lots of flags, but if youâve got some really common ones, display them first. For example, the Git command displays the commands for getting started and the most commonly used subcommands first:
$ git
usage: git [--version] [--help] [-C <path>] [-c <name>=<value>]
[--exec-path[=<path>]] [--html-path] [--man-path] [--info-path]
[-p | --paginate | -P | --no-pager] [--no-replace-objects] [--bare]
[--git-dir=<path>] [--work-tree=<path>] [--namespace=<name>]
<command> [<args>]
These are common Git commands used in various situations:
start a working area (see also: git help tutorial)
clone Clone a repository into a new directory
init Create an empty Git repository or reinitialize an existing one
work on the current change (see also: git help everyday)
add Add file contents to the index
mv Move or rename a file, a directory, or a symlink
reset Reset current HEAD to the specified state
rm Remove files from the working tree and from the index
examine the history and state (see also: git help revisions)
bisect Use binary search to find the commit that introduced a bug
grep Print lines matching a pattern
log Show commit logs
show Show various types of objects
status Show the working tree status
âŠ
Use formatting in your help text. Bold headings make it much easier to scan. But, try to do it in a terminal-independent way so that your users aren't staring down a wall of escape characters.
$ heroku apps --help
list your apps
USAGE
$ heroku apps
OPTIONS
-A, --all include apps in all teams
-p, --personal list apps in personal account when a default team is set
-s, --space=space filter by space
-t, --team=team team to use
--json output in json format
EXAMPLES
$ heroku apps
=== My Apps
example
example2
=== Collaborated Apps
theirapp other@owner.name
COMMANDS
apps:create creates a new app
apps:destroy permanently destroy an app
apps:errors view app errors
apps:favorites list favorited apps
apps:info show detailed app information
apps:join add yourself to a team app
apps:leave remove yourself from a team app
apps:lock prevent team members from joining an app
apps:open open the app in a web browser
apps:rename rename an app
apps:stacks show the list of available stacks
apps:transfer transfer applications to another user or team
apps:unlock unlock an app so any team member can join
Note: When heroku apps --help is piped through a pager, the command emits no escape characters.
If the user did something wrong and you can guess what they meant, suggest it.
For example, brew update jq tells you that you should run brew upgrade jq.
You can ask if they want to run the suggested command, but donât force it on them. For example:
$ heroku pss
âș Warning: pss is not a heroku command.
Did you mean ps? [y/n]:
Rather than suggesting the corrected syntax, you might be tempted to just run it for them, as if theyâd typed it right in the first place. Sometimes this is the right thing to do, but not always.
Firstly, invalid input doesnât necessarily imply a simple typoâit can often mean the user has made a logical mistake, or misused a shell variable. Assuming what they meant can be dangerous, especially if the resulting action modifies state.
Secondly, be aware that if you change what the user typed, they wonât learn the correct syntax. In effect, youâre ruling that the way they typed it is valid and correct, and youâre committing to supporting that indefinitely. Be intentional in making that decision, and document both syntaxes.
Further reading: âDo What I Meanâ
If your command is expecting to have something piped to it and stdin is an interactive terminal, display help immediately and quit.
This means it doesnât just hang, like cat.
Alternatively, you could print a log message to stderr.
The purpose of help text is to give a brief, immediate sense of what your tool is, what options are available, and how to perform the most common tasks. Documentation, on the other hand, is where you go into full detail. Itâs where people go to understand what your tool is for, what it isnât for, how it works and how to do everything they might need to do.
Provide web-based documentation. People need to be able to search online for your toolâs documentation, and to link other people to specific parts. The web is the most inclusive documentation format available.
Provide terminal-based documentation. Documentation in the terminal has several nice properties: itâs fast to access, it stays in sync with the specific installed version of the tool, and it works without an internet connection.
Consider providing man pages.
man pages, Unixâs original system of documentation, are still in use today, and many users will reflexively check man mycmd as a first step when trying to learn about your tool.
To make them easier to generate, you can use a tool like ronn (which can also generate your web docs).
However, not everyone knows about man, and it doesnât run on all platforms, so you should also make sure your terminal docs are accessible via your tool itself.
For example, git and npm make their man pages accessible via the help subcommand, so npm help ls is equivalent to man npm-ls.
NPM-LS(1) NPM-LS(1)
NAME
npm-ls - List installed packages
SYNOPSIS
npm ls [[<@scope>/]<pkg> ...]
aliases: list, la, ll
DESCRIPTION
This command will print to stdout all the versions of packages that are
installed, as well as their dependencies, in a tree-structure.
...
Human-readable output is paramount.
Humans come first, machines second.
The most simple and straightforward heuristic for whether a particular output stream (stdout or stderr) is being read by a human is whether or not itâs a TTY.
Whatever language youâre using, it will have a utility or library for doing this (e.g. Python, Node, Go).
Further reading on what a TTY is.
Have machine-readable output where it does not impact usability.
Streams of text is the universal interface in UNIX.
Programs typically output lines of text, and programs typically expect lines of text as input,
therefore you can compose multiple programs together.
This is normally done to make it possible to write scripts,
but it can also help the usability for humans using programs.
For example, a user should be able to pipe output to grep and it should do what they expect.
âExpect the output of every program to become the input to another, as yet unknown, program.â â Doug McIlroy
If human-readable output breaks machine-readable output, use --plain to display output in plain, tabular text format for integration with tools like grep or awk.
In some cases, you might need to output information in a different way to make it human-readable.
For example, if you are displaying a line-based table, you might choose to split a cell into multiple lines, fitting in more information while keeping it within the width of the screen.
This breaks the expected behavior of there being one piece of data per line, so you should provide a --plain flag for scripts, which disables all such manipulation and outputs one record per line.
Display output as formatted JSON if --json is passed.
JSON allows for more structure than plain text, so it makes it much easier to output and handle complex data structures.
jq is a common tool for working with JSON on the command-line, and there is now a whole ecosystem of tools that output and manipulate JSON.
It is also widely used on the web, so by using JSON as the input and output of programs, you can pipe directly to and from web services using curl.
Display output on success, but keep it brief.
Traditionally, when nothing is wrong, UNIX commands display no output to the user.
This makes sense when theyâre being used in scripts, but can make commands appear to be hanging or broken when used by humans.
For example, cp will not print anything, even if it takes a long time.
Itâs rare that printing nothing at all is the best default behavior, but itâs usually best to err on the side of less.
For instances where you do want no output (for example, when used in shell scripts), to avoid clumsy redirection of stderr to /dev/null, you can provide a -q option to suppress all non-essential output.
If you change state, tell the user. When a command changes the state of a system, itâs especially valuable to explain what has just happened, so the user can model the state of the system in their headâparticularly if the result doesnât directly map to what the user requested.
For example, git push tells you exactly what it is doing, and what the new state of the remote branch is:
$ git push
Enumerating objects: 18, done.
Counting objects: 100% (18/18), done.
Delta compression using up to 8 threads
Compressing objects: 100% (10/10), done.
Writing objects: 100% (10/10), 2.09 KiB | 2.09 MiB/s, done.
Total 10 (delta 8), reused 0 (delta 0), pack-reused 0
remote: Resolving deltas: 100% (8/8), completed with 8 local objects.
To github.com:replicate/replicate.git
+ 6c22c90...a2a5217 bfirsh/fix-delete -> bfirsh/fix-delete
Make it easy to see the current state of the system. If your program does a lot of complex state changes and it is not immediately visible in the filesystem, make sure you make this easy to view.
For example, git status tells you as much information as possible about the current state of your Git repository, and some hints at how to modify the state:
$ git status
On branch bfirsh/fix-delete
Your branch is up to date with 'origin/bfirsh/fix-delete'.
Changes not staged for commit:
(use "git add <file>..." to update what will be committed)
(use "git restore <file>..." to discard changes in working directory)
modified: cli/pkg/cli/rm.go
no changes added to commit (use "git add" and/or "git commit -a")
Suggest commands the user should run.
When several commands form a workflow, suggesting to the user commands they can run next helps them learn how to use your program and discover new functionality.
For example, in the git status output above, it suggests commands you can run to modify the state you are viewing.
Actions crossing the boundary of the programâs internal world should usually be explicit. This includes things like:
Increase information densityâwith ASCII art!
For example, ls shows permissions in a scannable way.
When you first see it, you can ignore most of the information.
Then, as you learn how it works, you pick out more patterns over time.
-rw-r--r-- 1 root root 68 Aug 22 23:20 resolv.conf
lrwxrwxrwx 1 root root 13 Mar 14 20:24 rmt -> /usr/sbin/rmt
drwxr-xr-x 4 root root 4.0K Jul 20 14:51 security
drwxr-xr-x 2 root root 4.0K Jul 20 14:53 selinux
-rw-r----- 1 root shadow 501 Jul 20 14:44 shadow
-rw-r--r-- 1 root root 116 Jul 20 14:43 shells
drwxr-xr-x 2 root root 4.0K Jul 20 14:57 skel
-rw-r--r-- 1 root root 0 Jul 20 14:43 subgid
-rw-r--r-- 1 root root 0 Jul 20 14:43 subuid
Use color with intention. For example, you might want to highlight some text so the user notices it, or use red to indicate an error. Donât overuse itâif everything is a different color, then the color means nothing and only makes it harder to read.
Disable color if your program is not in a terminal or the user requested it. These things should disable colors:
stdout or stderr is not an interactive terminal (a TTY).
Itâs best to individually checkâif youâre piping stdout to another program, itâs still useful to get colors on stderr.NO_COLOR environment variable is set and it is not empty (regardless of its value).TERM environment variable has the value dumb.--no-color.MYAPP_NO_COLOR environment variable in case users want to disable color specifically for your program.Further reading: no-color.org, 12 Factor CLI Apps
If stdout is not an interactive terminal, donât display any animations.
This will stop progress bars turning into Christmas trees in CI log output.
Use symbols and emoji where it makes things clearer. Pictures can be better than words if you need to make several things distinct, catch the userâs attention, or just add a bit of character. Be careful, thoughâit can be easy to overdo it and make your program look cluttered or feel like a toy.
For example, yubikey-agent uses emoji to add structure to the output so it isnât just a wall of text, and a â to draw your attention to an important piece of information:
$ yubikey-agent -setup
đ The PIN is up to 8 numbers, letters, or symbols. Not just numbers!
â The key will be lost if the PIN and PUK are locked after 3 incorrect tries.
Choose a new PIN/PUK:
Repeat the PIN/PUK:
đ§Ș Retriculating splines âŠ
â
Done! This YubiKey is secured and ready to go.
đ€ When the YubiKey blinks, touch it to authorize the login.
đ Here's your new shiny SSH public key:
ecdsa-sha2-nistp256 AAAAE2VjZHNhLXNoYTItbmlzdHAyNTYAAAAIbmlzdHAyNTYAAABBBCEJ/
UwlHnUFXgENO3ifPZd8zoSKMxESxxot4tMgvfXjmRp5G3BGrAnonncE7Aj11pn3SSYgEcrrn2sMyLGpVS0=
đ Remember: everything breaks, have a backup plan for when this YubiKey does.
By default, donât output information thatâs only understandable by the creators of the software. If a piece of output serves only to help you (the developer) understand what your software is doing, it almost certainly shouldnât be displayed to normal users by defaultâonly in verbose mode.
Invite usability feedback from outsiders and people who are new to your project. Theyâll help you see important issues that you are too close to the code to notice.
Donât treat stderr like a log file, at least not by default.
Donât print log level labels (ERR, WARN, etc.) or extraneous contextual information, unless in verbose mode.
Use a pager (e.g. less) if you are outputting a lot of text.
For example, git diff does this by default.
Using a pager can be error-prone, so be careful with your implementation such that you donât make the experience worse for the user.
Use a pager only if stdin or stdout is an interactive terminal.
A good sensible set of options to use for less is less -FIRX.
This does not page if the content fills one screen, ignores case when you search, enables color and formatting, and leaves the contents on the screen when less quits.
There might be libraries in your language that are more robust than piping to less.
For example, pypager in Python.
One of the most common reasons to consult documentation is to fix errors. If you can make errors into documentation, then this will save the user loads of time.
Catch errors and rewrite them for humans. If youâre expecting an error to happen, catch it and rewrite the error message to be useful. Think of it like a conversation, where the user has done something wrong and the program is guiding them in the right direction. Example: âCanât write to file.txt. You might need to make it writable by running âchmod +w file.txtâ.â
Signal-to-noise ratio is crucial. The more irrelevant output you produce, the longer itâs going to take the user to figure out what they did wrong. If your program produces multiple errors of the same type, consider grouping them under a single explanatory header instead of printing many similar-looking lines.
Consider where the user will look first. Put the most important information at the end of the output. The eye will be drawn to red text, so use it intentionally and sparingly.
If there is an unexpected or unexplainable error, provide debug and traceback information, and instructions on how to submit a bug. That said, donât forget about the signal-to-noise ratio: you donât want to overwhelm the user with information they donât understand. Consider writing the debug log to a file instead of printing it to the terminal.
Make it effortless to submit bug reports. One nice thing you can do is provide a URL and have it pre-populate as much information as possible.
Further reading: Google: Writing Helpful Error Messages, Nielsen Norman Group: Error-Message Guidelines
A note on terminology:
cp are args.
The order of args is often important: cp foo bar means something different from cp bar foo.-r) or a double hyphen and a multiple-letter name (--recursive).
They may or may not also include a user-specified value (--file foo.txt, or --file=foo.txt).
The order of flags, generally speaking, does not affect program semantics.Prefer flags to args. Itâs a bit more typing, but it makes it much clearer what is going on. It also makes it easier to make changes to how you accept input in the future. Sometimes when using args, itâs impossible to add new input without breaking existing behavior or creating ambiguity.
Citation: 12 Factor CLI Apps.
Have full-length versions of all flags.
For example, have both -h and --help.
Having the full version is useful in scripts where you want to be verbose and descriptive, and you donât have to look up the meaning of flags everywhere.
Citation: GNU Coding Standards.
Only use one-letter flags for commonly used flags, particularly at the top-level when using subcommands. That way you donât âpolluteâ your namespace of short flags, forcing you to use convoluted letters and cases for flags you add in the future.
Multiple arguments are fine for simple actions against multiple files.
For example, rm file1.txt file2.txt file3.txt.
This also makes it work with globbing: rm *.txt.
If youâve got two or more arguments for different things, youâre probably doing something wrong.
The exception is a common, primary action, where the brevity is worth memorizing.
For example, cp <source> <destination>.
Citation: 12 Factor CLI Apps.
Use standard names for flags, if there is a standard. If another commonly used command uses a flag name, itâs best to follow that existing pattern. That way, a user doesnât have to remember two different options (and which command it applies to), and users can even guess an option without having to look at the help text.
Here's a list of commonly used options:
-a, --all: All.
For example, ps, fetchmail.-d, --debug: Show debugging output.-f, --force: Force.
For example, rm -f will force the removal of files, even if it thinks it does not have permission to do it.
This is also useful for commands which are doing something destructive that usually require user confirmation, but you want to force it to do that destructive action in a script.--json: Display JSON output.
See the output section.-h, --help: Help.
This should only mean help.
See the help section.-n, --dry-run: Dry run.
Do not run the command, but describe the changes that would occur if the command were run. For example, rsync, git add.--no-input: See the interactivity section.-o, --output: Output file.
For example, sort, gcc.-p, --port: Port.
For example, psql, ssh.-q, --quiet: Quiet.
Display less output.
This is particularly useful when displaying output for humans that you might want to hide when running in a script.-u, --user: User.
For example, ps, ssh.--version: Version.-v: This can often mean either verbose or version.
You might want to use -d for verbose and this for version, or for nothing to avoid confusion.Make the default the right thing for most users. Making things configurable is good, but most users are not going to find the right flag and remember to use it all the time (or alias it). If itâs not the default, youâre making the experience worse for most of your users.
For example, ls has terse default output to optimize for scripts and other historical reasons, but if it were designed today, it would probably default to ls -lhF.
Prompt for user input. If a user doesnât pass an argument or flag, prompt for it. (See also: Interactivity)
Never require a prompt.
Always provide a way of passing input with flags or arguments.
If stdin is not an interactive terminal, skip prompting and just require those flags/args.
Confirm before doing anything dangerous.
A common convention is to prompt for the user to type y or yes if running interactively, or requiring them to pass -f or --force otherwise.
âDangerousâ is a subjective term, and there are differing levels of danger:
--confirm="name-of-thing", so itâs still scriptable.Consider whether there are non-obvious ways to accidentally destroy things. For example, imagine a situation where changing a number in a configuration file from 10 to 1 means that 9 things will be implicitly deletedâthis should be considered a severe risk, and should be difficult to do by accident.
If input or output is a file, support - to read from stdin or write to stdout.
This lets the output of another command be the input of your command and vice versa, without using a temporary file.
For example, tar can extract files from stdin:
$ curl https://example.com/something.tar.gz | tar xvf -
If a flag can accept an optional value, allow a special word like ânoneâ.
For example, ssh -F takes an optional filename of an alternative ssh_config file, and ssh -F none runs SSH with no config file. Donât just use a blank valueâthis can make it ambiguous whether arguments are flag values or arguments.
If possible, make arguments, flags and subcommands order-independent.
A lot of CLIs, especially those with subcommands, have unspoken rules on where you can put various arguments.
For example a command might have a --foo flag that only works if you put it before the subcommand:
mycmd --foo=1 subcmd
works
$ mycmd subcmd --foo=1
unknown flag: --foo
This can be very confusing for the userâespecially given that one of the most common things users do when trying to get a command to work is to hit the up arrow to get the last invocation, stick another option on the end, and run it again. If possible, try to make both forms equivalent, although you might run up against the limitations of your argument parser.
Do not read secrets directly from flags.
When a command accepts a secret, e.g. via a --password flag,
the flag value will leak the secret into ps output and potentially shell history.
And, this sort of flag encourages the use of insecure environment variables for secrets.
(Environment variables are insecure because they can often be read by other users, their values end up in debug logs, etc.)
Consider accepting sensitive data only via files, e.g. with a --password-file flag, or via stdin.
A --password-file flag allows a secret to be passed in discreetly, in a wide variety of contexts.
(Itâs possible to pass a fileâs contents into a flag in Bash by using --password $(< password.txt).
This approach has the same security problems mentioned above.
Itâs best avoided.)
Only use prompts or interactive elements if stdin is an interactive terminal (a TTY).
This is a pretty reliable way to tell whether youâre piping data into a command or whether it's being run in a script, in which case a prompt wonât work and you should throw an error telling the user what flag to pass.
If --no-input is passed, donât prompt or do anything interactive.
This allows users an explicit way to disable all prompts in commands.
If the command requires input, fail and tell the user how to pass the information as a flag.
If youâre prompting for a password, donât print it as the user types. This is done by turning off echo in the terminal. Your language should have helpers for this.
Let the user escape.
Make it clear how to get out.
(Donât do what vim does.)
If your program hangs on network I/O etc, always make Ctrl-C still work.
If itâs a wrapper around program execution where Ctrl-C canât quit (SSH, tmux, telnet, etc), make it clear how to do that.
For example, SSH allows escape sequences with the ~ escape character.
If youâve got a tool thatâs sufficiently complex, you can reduce its complexity by making a set of subcommands. If you have several tools that are very closely related, you can make them easier to use and discover by combining them into a single command (for example, RCS vs. Git).
Theyâre useful for sharing stuffâglobal flags, help text, configuration, storage mechanisms.
Be consistent across subcommands. Use the same flag names for the same things, have similar output formatting, etc.
Use consistent names for multiple levels of subcommand.
If a complex piece of software has lots of objects and operations that can be performed on those objects, it is a common pattern to use two levels of subcommand for this, where one is a noun and one is a verb.
For example, docker container create.
Be consistent with the verbs you use across different types of objects.
Either noun verb or verb noun ordering works, but noun verb seems to be more common.
Further reading: User experience, CLIs, and breaking the world, by John Starich.
Donât have ambiguous or similarly-named commands. For example, having two subcommands called âupdateâ and âupgradeâ is quite confusing. You might want to use different words, or disambiguate with extra words.
Validate user input. Everywhere your program accepts data from the user, it will eventually be given bad data. Check early and bail out before anything bad happens, and make the errors understandable.
Responsive is more important than fast. Print something to the user in <100ms. If youâre making a network request, print something before you do it so it doesnât hang and look broken.
Show progress if something takes a long time. If your program displays no output for a while, it will look broken. A good spinner or progress indicator can make a program appear to be faster than it is.
Ubuntu 20.04 has a nice progress bar that sticks to the bottom of the terminal.
If the progress bar gets stuck in one place for a long time, the user wonât know if stuff is still happening or if the programâs crashed. Itâs good to show estimated time remaining, or even just have an animated component, to reassure them that youâre still working on it.
There are many good libraries for generating progress bars. For example, tqdm for Python, schollz/progressbar for Go, and node-progress for Node.js.
Do stuff in parallel where you can, but be thoughtful about it. Itâs already difficult to report progress in the shell; doing it for parallel processes is ten times harder. Make sure itâs robust, and that the output isnât confusingly interleaved. If you can use a library, do soâthis is code you donât want to write yourself. Libraries like tqdm for Python and schollz/progressbar for Go support multiple progress bars natively.
The upside is that it can be a huge usability gain.
For example, docker pullâs multiple progress bars offer crucial insight into whatâs going on.
$ docker image pull ruby
Using default tag: latest
latest: Pulling from library/ruby
6c33745f49b4: Pull complete
ef072fc32a84: Extracting [================================================> ] 7.569MB/7.812MB
c0afb8e68e0b: Download complete
d599c07d28e6: Download complete
f2ecc74db11a: Downloading [=======================> ] 89.11MB/192.3MB
3568445c8bf2: Download complete
b0efebc74f25: Downloading [===========================================> ] 19.88MB/22.88MB
9cb1ba6838a0: Download complete
One thing to be aware of: hiding logs behind progress bars when things go well makes it much easier for the user to understand whatâs going on, but if there is an error, make sure you print out the logs. Otherwise, it will be very hard to debug.
Make things time out. Allow network timeouts to be configured, and have a reasonable default so it doesnât hang forever.
Make it recoverable.
If the program fails for some transient reason (e.g. the internet connection went down), you should be able to hit <up> and <enter> and it should pick up from where it left off.
Make it crash-only. This is the next step up from idempotence. If you can avoid needing to do any cleanup after operations, or you can defer that cleanup to the next run, your program can exit immediately on failure or interruption. This makes it both more robust and more responsive.
Citation: Crash-only software: More than meets the eye.
People are going to misuse your program. Be prepared for that. They will wrap it in scripts, use it on bad internet connections, run many instances of it at once, and use it in environments you havenât tested in, with quirks you didnât anticipate. (Did you know macOS filesystems are case-insensitive but also case-preserving?)
In software of any kind, itâs crucial that interfaces donât change without a lengthy and well-documented deprecation process. Subcommands, arguments, flags, configuration files, environment variables: these are all interfaces, and youâre committing to keeping them working. (Semantic versioning can only excuse so much change; if youâre putting out a major version bump every month, itâs meaningless.)
Keep changes additive where you can. Rather than modify the behavior of a flag in a backwards-incompatible way, maybe you can add a new flagâas long as it doesnât bloat the interface too much. (See also: Prefer flags to args.)
Warn before you make a non-additive change. Eventually, youâll find that you canât avoid breaking an interface. Before you do, forewarn your users in the program itself: when they pass the flag youâre looking to deprecate, tell them itâs going to change soon. Make sure thereâs a way they can modify their usage today to make it future-proof, and tell them how to do it.
If possible, you should detect when theyâve changed their usage and not show the warning any more: now they wonât notice a thing when you finally roll out the change.
Changing output for humans is usually OK.
The only way to make an interface easy to use is to iterate on it, and if the output is considered an interface, then you canât iterate on it.
Encourage your users to use --plain or --json in scripts to keep output stable (see Output).
Donât have a catch-all subcommand.
If you have a subcommand thatâs likely to be the most-used one, you might be tempted to let people omit it entirely for brevityâs sake.
For example, say you have a run command that wraps an arbitrary shell command:
$ mycmd run echo "hello world"
You could make it so that if the first argument to mycmd isnât the name of an existing subcommand, you assume the user means run, so they can just type this:
$ mycmd echo "hello world"
This has a serious drawback, though: now you can never add a subcommand named echoâor anything at allâwithout risking breaking existing usages.
If thereâs a script out there that uses mycmd echo, it will do something entirely different after that user upgrades to the new version of your tool.
Donât allow arbitrary abbreviations of subcommands.
For example, say your command has an install subcommand.
When you added it, you wanted to save users some typing, so you allowed them to type any non-ambiguous prefix, like mycmd ins, or even just mycmd i, and have it be an alias for mycmd install.
Now youâre stuck: you canât add any more commands beginning with i, because there are scripts out there that assume i means install.
Thereâs nothing wrong with aliasesâsaving on typing is goodâbut they should be explicit and remain stable.
Donât create a âtime bomb.â Imagine itâs 20 years from now. Will your command still run the same as it does today, or will it stop working because some external dependency on the internet has changed or is no longer maintained? The server most likely to not exist in 20 years is the one that you are maintaining right now. (But donât build in a blocking call to Google Analytics either.)
If a user hits Ctrl-C (the INT signal), exit as soon as possible. Say something immediately, before you start clean-up. Add a timeout to any clean-up code so it canât hang forever.
If a user hits Ctrl-C during clean-up operations that might take a long time, skip them. Tell the user what will happen when they hit Ctrl-C again, in case it is a destructive action.
For example, when quitting Docker Compose, you can hit Ctrl-C a second time to force your containers to stop immediately instead of shutting them down gracefully.
$ docker-compose up
âŠ
^CGracefully stopping... (press Ctrl+C again to force)
Your program should expect to be started in a situation where clean-up has not been run. (See Crash-only software: More than meets the eye.)
Command-line tools have lots of different types of configuration, and lots of different ways to supply it (flags, environment variables, project-level config files). The best way to supply each piece of configuration depends on a few factors, chief among them specificity, stability and complexity.
Configuration generally falls into a few categories:
Likely to vary from one invocation of the command to the next.
Examples:
Recommendation: Use flags. Environment variables may or may not be useful as well.
Generally stable from one invocation to the next, but not always. Might vary between projects. Definitely varies between different users working on the same project.
This type of configuration is often specific to an individual computer.
Examples:
Recommendation: Use flags and probably environment variables too.
Users may want to set the variables in their shell profile so they apply globally, or in .env for a particular project.
If this configuration is sufficiently complex, it may warrant a configuration file of its own, but environment variables are usually good enough.
Stable within a project, for all users.
This is the type of configuration that belongs in version control.
Files like Makefile, package.json and docker-compose.yml are all examples of this.
Recommendation: Use a command-specific, version-controlled file.
Follow the XDG-spec.
In 2010 the X Desktop Group, now freedesktop.org, developed a specification for the location of base directories where config files may be located.
One goal was to limit the proliferation of dotfiles in a userâs home directory by supporting a general-purpose ~/.config folder.
The XDG Base Directory Specification (full spec, summary) is supported by yarn, fish, wireshark, emacs, neovim, tmux, and many other projects you know and love.
If you automatically modify configuration that is not your programâs, ask the user for consent and tell them exactly what youâre doing.
Prefer creating a new config file (e.g. /etc/cron.d/myapp) rather than appending to an existing config file (e.g. /etc/crontab).
If you have to append or modify to a system-wide config file, use a dated comment in that file to delineate your additions.
Apply configuration parameters in order of precedence. Here is the precedence for config parameters, from highest to lowest:
.env)Environment variables are for behavior that varies with the context in which a command is run. The âenvironmentâ of an environment variable is the terminal sessionâthe context in which the command is running. So, an env var might change each time a command runs, or between terminal sessions on one machine, or between instantiations of one project across several machines.
Environment variables may duplicate the functionality of flags or configuration parameters, or they may be distinct from those things. See Configuration for a breakdown of common types of configuration and recommendations on when environment variables are most appropriate.
For maximum portability, environment variable names must only contain uppercase letters, numbers, and underscores (and mustn't start with a number).
Which means O_O and OWO are the only emoticons that are also valid environment variable names.
Aim for single-line environment variable values.
While multi-line values are possible, they create usability issues with the env command.
Avoid commandeering widely used names. Hereâs a list of POSIX standard env vars.
Check general-purpose environment variables for configuration values when possible:
NO_COLOR, to disable color (see Output) or FORCE_COLOR to enable it and ignore the detection logicDEBUG, to enable more verbose outputEDITOR, if you need to prompt the user to edit a file or input more than a single lineHTTP_PROXY, HTTPS_PROXY, ALL_PROXY and NO_PROXY, if youâre going to perform network operations
(The HTTP library youâre using might already check for these.)SHELL, if you need to open up an interactive session of the user's preferred shell
(If you need to execute a shell script, use a specific interpreter like /bin/sh)TERM, TERMINFO and TERMCAP, if youâre going to use terminal-specific escape sequencesTMPDIR, if youâre going to create temporary filesHOME, for locating configuration filesPAGER, if you want to automatically page outputLINES and COLUMNS, for output thatâs dependent on screen size (e.g. tables)Read environment variables from .env where appropriate.
If a command defines environment variables that are unlikely to change as long as the user is working in a particular directory,
then it should also read them from a local .env file so users can configure it differently for different projects without having to specify them every time.
Many languages have libraries for reading .env files (Rust, Node, Ruby).
Donât use .env as a substitute for a proper configuration file.
.env files have a lot of limitations:
.env file is not commonly stored in source controlIf it seems like these limitations will hamper usability or security, then a dedicated config file might be more appropriate.
Do not read secrets from environment variables. While environment variables may be convenient for storing secrets, they have proven too prone to leakage:
curl -H "Authorization: Bearer $BEARER_TOKEN" will leak into globally-readable process state.
(cURL offers the -H @filename alternative for reading sensitive headers from a file.)docker inspectsystemctl showSecrets should only be accepted via credential files, pipes, AF_UNIX sockets, secret management services, or another IPC mechanism.
âNote the obsessive use of abbreviations and avoidance of capital letters; [Unix] is a system invented by people to whom repetitive stress disorder is what black lung is to miners. Long names get worn down to three-letter nubbins, like stones smoothed by a river.â â Neal Stephenson, In the Beginning was the Command Line
The name of your program is particularly important on the CLI: your users will be typing it all the time, and it needs to be easy to remember and type.
Make it a simple, memorable word.
But not too generic, or youâll step on the toes of other commands and confuse users.
For example, both ImageMagick and Windows used the command convert.
Use only lowercase letters, and dashes if you really need to.
curl is a good name, DownloadURL is not.
Keep it short.
Users will be typing it all the time.
Donât make it too short: the very shortest commands are best reserved for the common utilities used all the time, such as cd, ls, ps.
Make it easy to type. If you expect people to type your command name all day, make it easy on their hands.
A real-world example: long before Docker Compose was docker compose, it was plum.
This turned out to be such an awkward, one-handed hopscotch that it was immediately renamed to fig, which â as well as being shorter â flows much more easily.
Further reading: The Poetics of CLI Command Names
If possible, distribute as a single binary. If your language doesnât compile to binary executables as standard, see if it has something like PyInstaller. If you really canât distribute as a single binary, use the platformâs native package installer so you arenât scattering things on disk that canât easily be removed. Tread lightly on the userâs computer.
If youâre making a language-specific tool, such as a code linter, then this rule doesnât applyâitâs safe to assume the user has an interpreter for that language installed on their computer.
Make it easy to uninstall. If it needs instructions, put them at the bottom of the install instructionsâone of the most common times people want to uninstall software is right after installing it.
Create and modify oclif-based Node.js CLIs and plugins. Use when scaffolding an oclif project, adding commands/flags/args, configuring topics/plugins/hooks/help, testing with @oclif/test, or packaging/releasing via npm, tarballs, or installers.
Use when working with the Paperless-ngx REST API; looking up endpoints, request/response shapes, auth requirements, filtering/pagination, or generating example requests/responses from the OpenAPI spec. Trigger for tasks involving Paperless-ngx API integrations, client code, curl examples, or schema details.